Proximal hyperspectral imaging has been widely used in material analysis, mineral mapping, and related geological studies. However, most publicly available datasets are limited to a single observation scenario and commonly lack corresponding mineralogical and geochemical constraints, which restricts quantitative analysis, cross-scenario comparison, and objective method evaluation. To meet this need, this study presents a multi-scenario proximal shortwave infrared imaging hyperspectral dataset for geological samples collected from typical clay-type lithium deposits hosted in carbonate successions in China, systematically covering four complementary observation scenarios, including powder, drill core, hand specimen, and in-situ outcrop. The data were acquired using a HySpex imaging spectrometer covering 960–2500 nm. In addition to the hyperspectral imagery, the dataset integrates mineralogical information, geochemical information, sampling annotations, reference panel spectra, geographic locations, and related metadata, thereby establishing a unified framework linking spectral observations, sample composition, and spatial context. For the controlled indoor scenarios, the dataset provides reflectance data, whereas for the field in-situ scenario it provides radiometrically corrected imagery together with reference panel spectra and example products for data use. The dataset is organized in standard and widely used formats. By covering multiple observation scenarios under the same geological setting and hyperspectral sensor system, and by providing mineralogical and geochemical reference information, it can support model development, domain-shift analysis, and studies linking proximal observations with multi-platform mineral mapping and spectral analysis. The dataset is publicly available on Zenodo at <a href="https://doi.org/10.5281/zenodo.19142250" target="_blank" rel="noopener">https://doi.org/10.5281/zenodo.19142250</a> (Zhang et al., 2026a).